Re: Package version numbering from upstream

2009-01-15 Thread Tim Spriggs

Hey Anil,

Oh I do hate this can of worms; it keeps getting opened/closed/re-opened.

--- explanation as it occurs to me at the moment, details may have 
drifted with memory ---


There is a hack within Nexenta's dpkg to understand any version ubuntuN 
= nexentaN even though this is wrong everywhere else. A long thread 
exists on this (maybe a year or two ago?) and essentially states that 
porting a package should  s/ubuntu/nexenta/. Things get complicated with 
upstream versions since we want a new upstream version to trump the 
nexenta version:


take the original upstream version:
1.2.3ubuntu1-1

which produced the nexenta version:
1.2.3nexenta1-1

but should be trumped by the next upstream revision
1.2.3ubuntu1-2  (should be higher)

this works AFAIK. As a packager you need to be weary of incrementing a 
version as is so tempting to do when patching an already in-repository 
source. Patching the version to:


1.2.3nexenta1-2

would make it as new as the upstream version:

1.2.3ubuntu1-2

This means we can easily lose sight of little upstream patches.


A slightly better way to increment the version is to add another -N to 
the end so that the version sequence would look like:


Orig:
1.2.3ubuntu1-1

Imported:
1.2.3nexenta1-1

Subsequent patches:
1.2.3nexenta1-1-1
1.2.3nexenta1-1-2
...
1.2.3nexenta1-1-N

This adds a lot of complexity to what a package maintainer must read up 
on, understand and remember for every package. Keeping track of this 
sort of thing seems error prone but would be as correct as you can get 
with version substitution as we know it.


The next sticking point is that versions in debian/control should be 
modified as well. Though things work as usual if you don't since the 
nexenta = ubuntu hack in dpkg exists.


Another thing to keep in mind is the ordering of source repositories in 
/etc/apt/sources.list when importing a source. You want the nexenta 
repositories first and then the ubuntu repositories so that the nexenta 
version that is the same as the ubuntu version is picked for download.


Why do we replace ubuntu with nexenta instead of appending nexenta1?

Great question, in theory this breaks some depends (when another package 
depends on lt, le, or eq) so we don't do it. In practice I have no idea 
how many depends this would actually break but I'm sure there would be 
at least a couple of packages. (This topic can start me on a rant abut 
not being able to increment internal versions anyway...)


--- /explanation ---

Alright, so anything ported over should be nexentaN instead of ubuntuN 
but the autobuilder does not make that change. This was a tough choice 
but the package that is imported by the ab is not modified in any 
significant way as to be considered ported. In this fashion it seemed 
to be wrong to consider the package as ported and so the source/binaries 
are left with the ubuntu version attached. If/when the autobuilder makes 
significant changes it should definitely change the version.



My personal opinion on the topic (for what that is worth) is that we 
should append nexentaN when we make a significant change and fix 
breakage elsewhere. It requires no special hacks in dpkg (or other 
related/semi-related tools) and is a lot more transparent to everyone 
involved (and those getting involved).


-Tim

Anil Gulecha wrote:

Hi,

This has come up a few times, so we should have a clear written policy
on how the version numbering changes from upstream into our
repository.

Lets take the following examples:
gtk-sharp2_2.12.0-2ubuntu3
zlib1g-1.2.3.3.dfsg-7ubuntu1
xyz-1.2.3.build2

The common practice is using apt-upstream-tool, which changes these as:
gtk-sharp2_2.12.0-2nexenta4
zlib1g-1.2.3.3.dfsg-7nexenta2
xyz-1.2.3.build2nexenta1

In some cases nexentaN is appended, in others [word]N is replaced with
nexentaN+1.

Autobuilder packages seem to keep the same version number.

We need a set of rules to fix these varied methods to something
consistent and logical. In certain cases, there are hardcoded version
rules in debian/control, ex: xyz ( 1.2.3ubuntu2) , which can fail
after version change to nexenta3.

A good plan seems to be to simply keep the upstream name, and add
nexenta1 , and increment for each of our revision. This keep things
like package versioning right (?).

Comments/suggestions on other ways to proceed?

Anil
  

___
gnusol-devel mailing list
gnusol-devel@lists.sonic.net
http://lists.sonic.net/mailman/listinfo/gnusol-devel


Top links to prstat

2009-01-15 Thread Jerome Warnier
Running NCP 2.0 up-to-date.

I noticed that top is a link to prstat, which is somewhat bad, as they
both don't have the same output or feature set (for example, prstat
supports zones, while top does not).

I guess the best solution would be to ship GNU top in NCP instead of
trying to simulate it.

Regards
___
gnusol-devel mailing list
gnusol-devel@lists.sonic.net
http://lists.sonic.net/mailman/listinfo/gnusol-devel


Re: Top links to prstat

2009-01-15 Thread Erast Benson
Is top ported to OpenSolaris yet? I'd like to see zone extensions in it
too... Also is anybody interested in porting GNU chmod and ls utilities
to work with ZFS ACL's ?

Such development, if successful, would somewhat close the gap between
SUN / GNU tools and make everybody happy, I think..

On Thu, 2009-01-15 at 10:12 +0100, Jerome Warnier wrote:
 Running NCP 2.0 up-to-date.
 
 I noticed that top is a link to prstat, which is somewhat bad, as they
 both don't have the same output or feature set (for example, prstat
 supports zones, while top does not).
 
 I guess the best solution would be to ship GNU top in NCP instead of
 trying to simulate it.
 
 Regards
 ___
 gnusol-devel mailing list
 gnusol-devel@lists.sonic.net
 http://lists.sonic.net/mailman/listinfo/gnusol-devel
 

___
gnusol-devel mailing list
gnusol-devel@lists.sonic.net
http://lists.sonic.net/mailman/listinfo/gnusol-devel


Re: Top links to prstat

2009-01-15 Thread David Bartley
On Thu, Jan 15, 2009 at 12:08 PM, Erast Benson er...@gnusolaris.org wrote:
 Is top ported to OpenSolaris yet? I'd like to see zone extensions in it
 too...

This top works: http://www.unixtop.org/. I don't think it's the same
one found on linux, but seems very close.

 Also is anybody interested in porting GNU chmod and ls utilities
 to work with ZFS ACL's ?

This is on my TODO list. There was limited discussion on the coreutils
list about this; the idea there was to have something platform
independent (apparently AIX and Mac-OS and others support some form of
non-POSIX ACL's too).

-- David


 Such development, if successful, would somewhat close the gap between
 SUN / GNU tools and make everybody happy, I think..

 On Thu, 2009-01-15 at 10:12 +0100, Jerome Warnier wrote:
 Running NCP 2.0 up-to-date.

 I noticed that top is a link to prstat, which is somewhat bad, as they
 both don't have the same output or feature set (for example, prstat
 supports zones, while top does not).

 I guess the best solution would be to ship GNU top in NCP instead of
 trying to simulate it.

 Regards
 ___
 gnusol-devel mailing list
 gnusol-devel@lists.sonic.net
 http://lists.sonic.net/mailman/listinfo/gnusol-devel


 ___
 gnusol-devel mailing list
 gnusol-devel@lists.sonic.net
 http://lists.sonic.net/mailman/listinfo/gnusol-devel


___
gnusol-devel mailing list
gnusol-devel@lists.sonic.net
http://lists.sonic.net/mailman/listinfo/gnusol-devel